For my habitat use study, I have used the method described by Byers (1984) and Neu et al. (1974) to determine habitat preference. I have managed to calculate the Chi Square, and would like to know how you determine whether the Chi Square test statistic is significant or not? For example, in Byers (1984), the Chi Square is 103.1, df=9, p
Hello Gaius,
There are lots of sources.
Here's an online calculator: https://homepage.divms.uiowa.edu/~mbognar/applets/chisq.html
Here's the command in excel: = 1 - CHISQ.DIST(chi, df, TRUE)
(where chi = the chi-square value; df = degrees of freedom)
Good luck with your work!
The R equivalent to David Morse 's answer follows. You can run R code at the following site without installing the software. https://rdrr.io/snippets/ .
1-pchisq(ch, df)
example, chi-square=20, df=5
1-pchisq(20, 5)
### p-value = 0.00125
### Same as Excel and uiowa calculator
Just for fun, I re-created the analysis in Neu et al. (1974). Note that the confidence intervals are somewhat different because I used a different method. It's not a bad idea to let the computer double-check your work. Again, the code can be run at the website cited above.
### Data from Neu et al. 1974
Observed = c(25, 22, 30, 40)
Sum = sum(Observed)
Expected = c(40, 12, 12, 53)/Sum
### Chi-square goodness-of-fit
chisq.test(Observed, p=Expected)
### Chi-squared test for given probabilities
### X-squared = 44.147, df = 3, p-value = 1.404e-09
### Proportions and confidence intervals
library(DescTools)
MultinomCI(Observed,
conf.level=0.95,
method="sisonglaz")
### est lwr.ci upr.ci
### [1,] 0.2136752 0.1282051 0.3160658
### [2,] 0.1880342 0.1025641 0.2904248
### [3,] 0.2564103 0.1709402 0.3588008
### [4,] 0.3418803 0.2564103 0.4442709
########################################
### Optional plot
### See: http://rcompanion.org/handbook/H_03.html
library(ggplot2)
Loc = c("In Burn", "Category 2", "Category3", "Out of Burn")
Obs = Observed/Sum
Location = c(Loc, Loc)
Location = factor(Location,
levels=unique(Location))
Proportion = c(rep("Observed", 4),
rep("Expected", 4))
Value = c(Obs, Expected)
DF = data.frame(Location, Proportion, Value)
DF
ggplot(DF,
aes(x = Location, y = Value, fill = Proportion)) +
geom_bar(stat = 'identity', position = 'dodge') +
scale_fill_manual(values = c("cornflowerblue", "blue")) +
ylab("Proportion")
Thank you very much David Morse , I did not know the answer was that simple. Very much appreciate your help. Best wishes.
Thank you Sal Mangiafico , I like how you have gone the extra length to recreate the Neu et al (1974) data, and the graph you have created. I have just over 60 such tables with different land use categories and other variables, for different animals (and hence the large number of tables). Is there a way to summarize these tables and/or create graphs exactly like the one you have created in R? Or would there be a need to show both graphs an tabular results? Thank you!
Gaius Wilson , there's always some way to use software to help summarize data, but how efficient this is depends on the complexity of the information to summarize and your proficiency at programming... Usually there's no need to present the same results both as a table and a plot, though it is sometimes useful. In the case of this kind of data, I thought the table in Neu et al. was effective. For the plot I presented, it would benefit from error bars on the observed values to indicate the confidence intervals.
- Badges
- Science topic